Google DeepMind Releases AlphaGenome 2 with New DNA Methylation Prediction

On November 11, 2026, Google DeepMind unveiled AlphaGenome 2, a next-generation DNA sequence model built to predict how DNA sequences and their variants affect specific human cell types and cell states. The company described it as the successor to the original AlphaGenome released in 2025, which focused on predicting how variants influence molecular activity at the level of whole tissues. Google DeepMind had previously also released the AlphaGenome Atlas, covering predictions for 9 billion single-base variants. The announcement was quickly picked up by The Verge, Techmeme, IT之家, cnBeta and other outlets.

According to IT之家 and The Verge, AlphaGenome 2 adopts a new conditional model architecture that takes the target tissue, cell type, or cell state as input, shifting predictions from tissue-average results down to single-cell resolution. Reports differ slightly on the training scale: Techmeme, citing The Verge, states the model was trained on more than 100 million cell profiles across 68 tissues, while IT之家 reports more than 107 million single-cell profiles from human and mouse samples, also covering 68 tissues. On performance, IT之家 says the new model identifies nearly twice as many gene-expression-altering variants as its predecessor, a figure The Decoder matches by reporting it finds nearly twice as many disease-relevant gene changes as the previous model. 智东西 adds that, alongside carrying over the core DNA functional assay metrics of the prior version, AlphaGenome 2 introduces DNA methylation prediction, which Google says covers hundreds of human cell types.

In the days after the launch, coverage focused on how researchers could access the model. MarkTechPost reports AlphaGenome 2 is available for non-commercial research through the AlphaGenome API, and The Decoder confirms the same access route while adding that the model is set to be integrated into Google's agentic platform Antigravity. That non-commercial release remains the current endpoint of the rollout.